Dubbing is not switched on yet. You can look around and create your account now.

Help

Jump to: Account · Dubbing & review · Plans · Privacy & safety · Troubleshooting. Not answered here? Contact us.

Account

I did not get the confirmation email

Check the spam folder, then press "Send the link again" on your projects page. Only the newest link works; it is valid for 24 hours.

I forgot my password

Use Forgot password. The link works once, for 30 minutes, and signs you out everywhere.

How do I delete my account?

Settings → Delete account. Your videos and dubs are deleted at once.

Dubbing & review

Which languages?

24, in any pair: Korean, English, Arabic, Danish, German, Greek, Spanish, Finnish, French, Hindi, Italian, Japanese, Malay, Dutch, Norwegian, Polish, Portuguese, Russian, Swedish, Swahili, Turkish, Chinese, Uzbek, Kazakh. Korean ↔ English is tested in depth; the others only on read speech (FLEURS) — see Results.

How long does a dub take?

On our GPU about 30 seconds per video minute into English and about 80 seconds into the other languages (their voice is slower), plus any wait in the queue. The progress page updates by itself.

What do the colours on the timeline mean?

Red: the translation is doubtful. Orange: the speech recognition is doubtful. Yellow: the dub does not fit its time even after speeding up. Green: no warning. Grey: flags hidden (evaluation mode).

Can I trust a green sentence?

Not fully. Flags catch doubtful sentences, but a translation can be wrong and confident. Listen to the whole dub before you publish it.

How do I correct a sentence?

Open it in the review editor, change the dub text, press Save, then Re-dub. Your text is never re-translated, and re-dubs use no minutes.

What are delivery tags?

[slow], [fast], [soft], [loud], [pause], [question], [whisper], [excited], [sad] and [angry] change how a sentence is spoken, with every voice: DUBBE changes the pitch, speed, loudness or breath itself. DUBBE also tags what it hears in the original — questions, emotion, laughs, sighs — shown dashed; untick the wrong ones. When you add [question] yourself, an English video switches to the Chatterbox voice, which raises the end of a question. [laugh], [sigh], [gasp] and [cough] are sounds only the English Chatterbox voice makes; other voices cannot, so the editor says "not performed".

Can every voice speak every language?

Yes. All 24 languages work with every preset voice, every clone of your own voice and every Marketplace voice: English is spoken by Kokoro, the other languages by Chatterbox, which takes the voice from a short recording. Hear any preset in any language under Voices. An English accent can carry over into other languages, more with some voices than others.

Can I get subtitles with a colour per speaker?

Yes — Subtitles makes them from any video: each speaker gets a colour (white, yellow, cyan, green …, the broadcast convention), as SRT, WebVTT and ASS, with an optional translation. Name the speakers, then burn the captions into the video if a platform cannot show caption files. For a dub: Download → Make the video with captions.

Several people speak in my video

DUBBE tells the speakers apart by voice and gives each a different voice. Under Speakers in the editor you can rename them, merge two that are one person, and in a sentence move it to another speaker or a new one. Similar voices can be taken for one person — check the coloured speaker chips.

Can a speaker keep their own voice?

Only with their consent: in the editor, "Use their own voice…" asks you to confirm the person agreed, then clones them from the video (at least 8 seconds of their speech). The copy stays in that project and is deleted with it. In Speaker Detection you can save it as a private voice instead; voices from videos can never be published.

My older project still has the old mistakes

Press Transcribe & translate again in the editor: the project runs again with the current recognition and translation models; your corrected sentences and tags are kept.

Can it use my own voice?

Yes. Under Voices → Clone my voice you read three sentences aloud in the browser; DUBBE checks that the recording says them and makes a clone that speaks every DUBBE language. Only your own voice, recorded live (no uploads); the recording is stored encrypted and deleted with the voice. Pick the clone in Text to Speech or in a dub's editor. You can share it in the voice library (an admin reviews it first) and stop sharing at any time.

What is "evaluation mode"?

It hides DUBBE's flags while you rate sentences, so your ratings are not influenced by them. We use these blind ratings to measure how good the flags are.

Plans

How many minutes do I have?

Free: 15 minutes a month, videos up to 10 minutes. See Plans.

Can I use dubs commercially?

Not during the beta: the licence review of every model is not finished, and the live Korean and Uzbek voices (MMS) are non-commercial.

Privacy & safety

Who can see my videos?

Only you (and, to fix problems, the DUBBE team). There are no public links.

When are my files deleted?

When you delete the project or your account, or automatically after your plan's retention period (Free: 7 days).

Is the dub marked as AI-generated?

Yes: in the file's metadata, and on the download page. Do not remove or hide that.

Troubleshooting

My upload was refused

Use MP4, MOV or MKV, at most 2 GB and within your plan's length and minutes. WebM and MKV are not accepted yet.

The job failed

The progress page shows the last lines of the log and a "Try again from this step" button. If it fails again, contact us with the project name.

The voice is the wrong gender or style

The voice is picked by pitch, which can go wrong with excited speech or several speakers. Choose another voice in the editor and re-dub.

The flags look wrong

Rate the sentence in the editor (what is wrong, and a score) — that is exactly the data we use to improve them.